Jingfeng Wu

Jingfeng Wu is a postdoctoral fellow at the Simons Institute for the Theory of Computing at UC Berkeley, hosted by Peter Bartlett and Bin Yu. He contributes to the NSF/Simons Collaboration on the Theoretical Foundations of Deep Learning, focusing on advancing core machine learning theory.

His academic journey includes a Ph.D. in Computer Science from Johns Hopkins University (2023), an M.S. in Applied Mathematics from Peking University (2019), and a B.S. in Mathematics from Peking University (2016). This strong mathematical foundation underpins his theoretical research approach.

Wu's research centers on deep learning theory, optimization algorithms, and statistical learning. He investigates implicit regularization in gradient descent, scaling laws in linear models, and the dynamics of large-stepsize optimization. His work reveals how non-monotonic loss landscapes and adaptive stepsizes accelerate convergence in logistic regression and deep networks, providing fundamental insights into generalization in overparameterized regimes.

Analysis of his 15 most recent publications shows a cohesive research program examining stochastic optimization in machine learning. Key themes include the interplay between stepsize selection and convergence rates, the role of data structure in scaling laws, and implicit bias mechanisms in neural networks. His work consistently bridges theoretical guarantees with practical algorithmic implications.

Scientific recognition includes:

• Rising Star in Data Science (2023) by University of Chicago and UC San Diego
• MINDS Summer Data Science Fellowship (2021)

Wu actively serves the research community as an organizer of the 2025 Deep Learning Theory Workshop at Simons Institute and as a reviewer for ICML (2020-2025), NeurIPS (2020-2025), and other top venues. His current work is supported by the NSF/Simons Collaboration grant.

He collaborates extensively with leading researchers including Peter Bartlett, Bin Yu, and Vladimir Braverman within the Simons Institute ecosystem, contributing to its vibrant theoretical machine learning community through seminars and collaborative projects.